Build scripts are the blind spot of AI coding: wrong builds waste CI minutes and nearly shipped a broken release
jimmykoppel · x · 2026-09-07
Engineer Jimmy Koppel shares how vibe coding burned him on build scripts, code he deliberately never reads because "if it builds, it works":
- A native dependency's prebuilt artifact was being built incorrectly, while the codebase already contained logic to rebuild it properly — silently wasting several minutes on every CI run
- A missing dependency in the build was masked by AI behavior: asked to "go build the thing," the agent just manually downloaded the package instead of fixing the script
He nearly shipped a bad release yesterday because bun.compile missed some packages, and now keeps a bevy of checks ensuring the compiled artifact bundles everything. His takeaway: for every part of the codebase allowed to be "vibed slop," it's still unclear whether skipping human attention actually saved time. The reply thread adds that for anything complex, reading the code beats running or testing it for understanding behavior — vibe coding fits only simple web apps without privacy or security requirements.
Related event: MIT researcher: reading code beats running it(4 posts)→
More from coding & agent
- Hands-on with ByteDance's Doubao Work client: delegate to cloud agents that finish while you walk away — AlchainHust · 2026-09-07
- Formo launches MCP server letting AI agents query onchain and product analytics — yosriady · 2026-09-07
- Nous Research's Hermes Desktop: free MIT-licensed native agent app for Mac, Windows and Linux — matviy · 2026-09-07
- Hermes Agent user slams 'endless bugs' in tools and integrations; Nous says he installed a community fork — diggz976 · 2026-09-07
- Dev Open-Sources Design of an LLM Memory Benchmark With Stale-Fact Scoring and Noise Scaling — True_Mongoose_7073 · 2026-09-07
- Nous Research's Teknium hits back at Hermes critic: repo has 17,500 open issues — Teknium · 2026-09-07